3rd SG13 Regional Workshop for Africa on “ITU-T Standardization Challenges
for Developing Countries Working for a Connected Africa”
(Livingstone, Zambia, 23-24 February 2015)
New Congestion Avoidance Methods during Planned
and Unplanned Failures for IP Network Architectures
Dr. Simon Tembo
(Dr.Eng., M.Eng., B.Eng., MIEEE, MEIZ, REng)
Assistant Dean Postgraduate Studies
School of Engineering
University of Zambia
simon.tembo@unza.zm
simontembo6@gmail.com
This Research was a Collaboration and Funded by Nippon
Telegraph & Telephone (NTT) Corp. of JAPAN.
PRESENTATION OUTLINE
1.
2.
3.
BACKGROUND
OBJECTIVES
PRESENT TWO CONGESTION AVOIDANCE METHODS:
(1) For Unplanned Failures
 A New Backup Topology Design Method for Congestion
Avoidance in IP Fast Reroute

International Journal of Networks and Communications 2012, Vol.2, No.5,
pp. 123-131, September 2012
(2) For Planned Failures
 A Method to Avoid Congestion during Transient Link Cost
Update in IP Networks

4.
5.
Plans to Submit for publication to the IEICE Transaction on Communications,
Japan
SUMMARY
ACKNOWLEDGEMENTS
2
BACKGROUND
 Due to increase in the use of Internet Backbones by carrying
Voice, Video and Data traffics, ISPs have SLAs with Customers to
guarantee Quality of Service (e.g. 99.999% Network Availability).
 Research shows that there are 2 Major Classes of Network
failures in IP Network Architecture [1] :
20%
Unplanned
Failures
Planned
Failures
80%
PLANNED FAILURES
UNPLANNED FAILURES
[1] : Characterization of Failures in an IP Backbone, by A. Markopoulou et. al , IEEE INFOCOM 2004
3
BACKGROUND
IP Network Failures
Unplanned Failures (80%)
Planned Failures (20%)
“Not Man made failures”
“Man made failures”
e.g. System or
Equipment Failure
e.g. Engineers shutting
down a Link for
Maintenance Routine
4
BACKGROUND
2 MAJOR OF CAUSES OF CONGESTION
5
OBJECTIVE
6
RESEARCH PARTS
1. Unplanned Failures : Present a new backup topology
design method to avoid congestion during IP Fast
Reroute (IPFRR):
 by using fewer backup topologies to reduce the size of
routing tables (i.e. reducing the router memory size)
2. Planned Failures : Present a link cost update method
to avoid congestion during transient link cost updates:
 Performing link cost updates is a Permutation
Problem;
 But selecting a congestion free link cost update during
transient period for Huge Permutations is a big
challenge for ISPs.
7
PART 1: CONGESTION AVOIDANCE FOR UNPLANNED FAILURE
A New Backup Topology Design Method for Congestion
Avoidance in IP Fast Reroute
1. INTRODUCTION
Existing Method: Multiple Routing Configurations (MRC)
Problem Statement
Research Originality
2. RESEARCH APPROACH
Overview
Algorithm
3. RESULTS & DISCUSSION
4. CONCLUSION
8
INTRODUCTION
TRADITIONAL IP NETWORKS – Backup path is calculated after Network Failure
Failure occurred
Fault Notification to
sent all routers
Computation process to
recalculate the new path
Re-Routing
IP FAST REROUTE METHOD – Backup path is calculated before Network Failure
Failure occurs
Re-Routing
Backup path is pre-arranged to
shortening the processing
In Traditional IP networks, failure recovery takes time.
IP Fast Reroute method, failure recovery is faster .
INTRODUCTION
EXAMPLE
Fault detected
After failure, fault
notice is sent to
all routers
Link Failure
2
1
After receiving the
failure notice, Router 1
then computes an
detour route
Fault detected
5
6
2
1
5
3
3
4
4
TRADITIONAL IP NETWORK
Detour Route
Pre-computed
6
IP FAST REROUTE NETWORK
10
INTRODUCTION
Multiple Routing Configurations (MRC) [2]
Original Topology
Backup Topology
Traffic is Re-routed
src
src
Fault detected
This link
is protected…
dst
Link Fails,
connectivity?
dst
Link sever has backup
topology pre-computed to
provide backup route for
continuous connectivity
when this link fails.
[2] A.Kvalbein, et.al, “Fast IP Network Recovery using Multiple Routing Configurations,” IEEE INFOCOM, Apr.2006.
11
INTRODUCTION
 Creating Many Backup Topologies -> NOT GOOD STRATEGY
 Because :1) It increases the size of routing tables kept in routers (Memory Problem).
2) It Increases link-state messages transmitted in the network.
3 Possible Backup Topologies for possible
network failures (with at most 10 Available Links)
Link Failure
src
Backup Topology #0
dst
Original Topology with
14 Available Links
Backup Topology #1
Backup Topology #2
Protected Link
12
INTRODUCTION – PROBLEM STATEMENT
 Reducing Number of Backup Topologies -> Fairly Good Strategy
We need to recover traffic with fewer backup
topologies for scalability purposes.
1. But reducing the number of backup topologies results
in limited available links for back up routes.
2. Due to fewer backup routes, rerouted traffic can
cause congestion if not carefully split among the
backup routes according to the available capacity.
13
INTRODUCTION – PROBLEM STATEMENT
• Problem with the existing approach (MRC) is the placement of
the fewer available links for backup routes; it lacks diverse routes
such that:
1) it causes congestion.
2) route optimization is not possible.
Backup Topology #0
is Selected
src
CONGESTION
3 Possible Backup Topologies for
all possible network failures
Backup Topology #0
1
2
4
6
dst
Original Topology
Backup Topology #1
3
5
7
8
Backup Topology #2
High-load node
Protected Link
14
14
RESEARCH APPROACH- RESEARCH ORIGINALITY

Proposed approach is to optimize the placement of protected links by:
1) Maximizing available links to overloaded nodes.
2) Splitting traffic on high load links to other links.
Proposed Approach
Existing Approach
Maximizing available
links to
Overloaded Nodes
High-load node
Protected Link
Special Node: Identify High-load node and
maximize the number of the available links
The key idea is the establishment of a SPECIAL NODE; i.e. Node with
1)High Traffic Volume - LOAD ORDER METHOD
2)High Node Degree – DEGREE ORDER METHOD
15
EXAMPLE OF SPECIAL NODE FOR LOAD ORDER & DEGREE ORDER
METHODS (HLDA NETWORK MODEL)
Top K Nodes = 3
(1) LOAD ORDER METHOD
1
Traffic Volume
9
2 METHODS OF DECIDING SPECIAL NODE
4
3
2
1000
800
600
400
200
0
1
2
3
4
5
6
7
Node ID
5
6
5
10
1
2
3
4
7
8
8
9
9
10
10
11
11
8
(2) DEGREE ORDER METHOD
11
10
7
2 METHODS OF SELECTING SPECIAL NODE
1) Top K method: Top-K influential nodes
(i.e. most influential nodes).
2) Swapping K method: Takes into
consideration the node position, when
selecting from the Top-K influential nodes
※Avoid selecting Neighbor node as
Special Node to that already chosen.
Node Degree
6
8
6
4
2
0
1
1
2
3
4
5 6 7
Node ID
8
9
2
3
4
5
6
9
10
7
8
10 11
11
RESEARCH APPROACH – ALGORITHM OVERVIEW
PROPOSED ALGORITHM STARTS BY:
1. Choosing the Special Node
2. Then maximize number of available links to the Special Node.
Existing Approach
Input
information
Proposed Approach
Topology
information
Topology information＋
Traffic Exchange
STEP1
Subdivision
All nodes will have
Link protection
arrangement
Selection of Special Node
STEP2
Special Nodes
Link protection arrangement
STEP3
Output Information
Backup Topology
Other nodes
Link protection arrangement
Backup Topology
17
COST 239 – Actual European Network Model (i.e. 11 nodes, 50 links)
The Evaluation involves quantitative measure of the link
load after a link failure
COPENHAGEN
6
LONDON
7
3
AMSTERDAM
5
BRUSSELS
8
BERLIN
LUXEMBOURG
2 PRAGUE
4
1
ZURICH
9
0
VIENNA
PARIS
10
MILAN
18
COST 266 – Actual European Network Model
(i.e. 26 nodes, 100 links)
The Evaluation involves quantitative measure
of the link load after a link failure
PERFORMANCE EVALUATION
 Evaluation Method
 Quantitative evaluation of the link load after a link failure
 Evaluated : Proposed Algorithm vs. Existing Algorithm
 Proposed : (Load Order Method and Degree Order Method)
 Conventional : Existing Method with no Special Node (K = 0)
 Evaluation Index
 Link load: total amount of traffic received in the link
 Evaluation Criteria
 Actual European Network Model
COST 239 (Smaller Network)
COST 266 (Larger Network)
 Traffic Model: Gravity model according to the population
distribution
20
RESEARCH RESULTS - LINK LOAD FOR COST 239 MODEL
(with 3 Backup Topologies )
Swapping K Method – For unplanned
failure able to Avoid Congestion
Top K Method – For unplanned
failure able to Avoid Congestion
Load Order
4.50
Load Order
4.50
75% Load Reduction
4.00
Degree Order
75% Load Reduction
4.00
Maximum Link Load
Maximum Link Load
Degree Order
3.50
3.00
2.50
2.00
1.50
1.00
0.50
3.50
3.00
2.50
2.00
1.50
1.00
0.50
0.00
0
1
2
3
4
Number of Special Nodes (K)
CONVENTIONAL METHOD
5
Conventional Method
Load Order Method
Degree Order Method
0.00
0
1
2
3
4
Number of Special Nodes (K)
5
CONGESTION FREE;
NORMAL CONDITION
21
RESULTS AND DISCUSSION-LINK LOAD FOR COST 266 MODEL
(with 4 Backup Topologies )
Swapping K Method – For unplanned
failure able to Avoid Congestion
Load Order
Load Order
Degree Order
33% Load Reduction
2.00
1.80
1.60
1.40
1.20
1.00
0.80
0.60
0.40
0.20
0.00
Maximum Link Load
Maximum Link Load
Top K Method –For unplanned
failure able to Reduce Congestion
0
1
2
3
4
Number of Special Nodes (K)
CONVENTIONAL METHOD
5
Conventional Method
Load Order Method
Degree Order Method
2.00
1.80
1.60
1.40
1.20
1.00
0.80
0.60
0.40
0.20
0.00
Degree Order
44% Load Reduction
0
1
2
3
4
Number of Special Nodes (K)
5
CONGESTION FREE;
NORMAL CONDITION
22
RESULTS AND DISCUSSION – COST 239 & COST266 MODELS
 The Proposed IPFRR method provides Congestion Avoidance
compared to the Existing IPFRR method (MRC method).
Network Model
Load Order Degree Order
(Traffic Volume) (Node Degree)
COST239
○
△
○
△
(Smaller)
COST 266
(Larger)
1. In both European Network Models: Load Order does Avoid
Congestion during unplanned failure than the Degree Order.
2. In Larger Networks: Load Order is only better than Degree
Order when node position of Special Nodes is considered.
DEFINITION OF SYMBOL
○： AVOID CONGESTION △： REDUCE CONGESTION
23
CONCLUSION – PART 1
1. Presented the Backup Topology design method to avoid
congestion for efficient IP Fast Reroute (IPFRR) during
unplanned failures.
2. The key idea of the method is the establishment of a Special
Node in a backup topology.
3. Using the proposed method we able to effect a load reduction
to about 75% as compared to existing method.
Paper:- S. Tembo, et. al. “A New Backup Topology Design Method for Congestion Avoidance in
IP Fast Reroute”, International Journal of Networks and Communications 2012, Vol.2, No.5,
pp. 123-131, September 2012, (http://www.sapub.org/journal/archive.aspx?journalid=1097).
Conference:- S. Tembo, et. al. “Dispersing Hotspot Traffic in Backup Topology for IP Fast
Reroute”, Proceedings of IEEE International Conference on Communications 2011
(IEEE ICC 2011), 5-9 June, Kyoto, Japan.
24
PART 2:- CONGESTION AVOIDANCE FOR PLANNED FAILURES
A Method to Avoid Congestion during Transient Link Cost
Update in IP Networks
1. INTRODUCTION
Example of Link Cost Update
Problem Statement
Research Originality
2. RESEARCH APPROACH
Overview
Algorithm
3. RESULTS & DISCUSSION
4. CONCLUSION
25
INTRODUCTION – EXAMPLE OF LINK COST UPDATE
 Example of link cost update from Initial Link Cost state to Target Link
Cost state for Traffic Engineering (TE) reason.
TARGET LINK COST STATE
INITIAL LINK COST STATE
High Load Link
High Load Link
1
2
1
1
src
1
3
1
1
1
4
1
2
2
6
2
5
Link 1-5 is
scheduled for
Maintenance
Routine
1
7
1
1
2
2
1
8
dst
1
5
3
1
1
4
1
2
2
1
6
2
2
5
10
7
1
8
Routing
Link Cost
There are 2 links needed to be updated.
Which link do we update 1st between
Link 1-5 & Link 3-7 to avoid congestion
in link 3-7?
26
INTRODUCTION – EXAMPLE OF LINK COST UPDATE
Performing link cost updates for several different
links is a Permutation Problem.
Example: For 3 different letters say A, B, C the
permutations are 6: i.e.
1. ABC
2. ACB
3. BAC
4. BCA
5. CAB
6. CBA
27
INTRODUCTION - PROBLEM STATEMENT
PROBLEM STATEMENT
Performing a link cost update for several different links for
TE purpose is a big challenge for Service Providers.
Finding a link cost update sequence order which is Congestion
free is not a practical computation time for Huge Permutations.
For Evaluation we use COST 239 (An European Network Model):
Present a method to avoid congestion free during
transient states.
TOPOLOGY
NODES
LINKS
COST 239
11
50
UPDATABLE
LINKS
6
PERMUTATIONS
720
HUGE PERMUTATIONS
28
PERFORMANCE EVALUATION
 EVALUATION METHOD
 Quantitative Evaluation of the Maximum Link Load
 EVALUATED : OUR ALGORITHMS VS. RANDOM ALGORITHM
 Random Method : Using Brute Force approach without applying the Our
Priority Link Cost Update Sequence Order (ULLN) method .
 Proposed : Using Brute Force [4] approach together with Our Priority Link
Cost Update Sequence Order (ULLN) method .
 EVALUATION INDEX
 Maximum Load in the Network
 EVALUATION CRITERIA
 Topology network model
COST 239 (A European Network Model)
 Traffic Model: Gravity Model according to the population distribution
 Configuration : Initial Link Cost & Target Link Cost
 DEFINITION OF OPTIMUM ROUTING AND CONGESTION
1) Optimum routing is when we have minimum maximum load in the Network.
2) Congestion occurs when the transient maximum load exceeds that of the
initial state.
29
[4] : Software Development by NTT Network Service Laboratories
COST 239 – A European Network Model
Which 6 links when updated gives Optimum Routing?
COPENHAGEN
6
LONDON
AMSTERDAM
7
5
BRUSSELS
8
3
BERLIN
LUXEMBOURG
4
2
1
ZURICH
9
PRAGUE
0
VIENNA
PARIS
10
MILAN
30
RANDOM UPDATE METHOD: USING BRUTE FORCE WITHOUT
APPLYING PROPOSED METHOD
Congestion occurs when the transient maximum load exceeds that of the initial state.
CONGESTION
0.88
Overflow
Threshold
MAXIMUM LOAD
0.86
0.84
0.82
0.8
0.78
0.76
INITIAL STATE
0.74
INITIAL
1)
2)
3)
1
2
3
4
5
TARGET
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
Congestion occurs during transient link cost update
Only 3 out of 24 patterns achieve optimum routing,(i.e. patterns 13, 16, & 22).
Optimum Routing Performance is 12.5% (3/24).
31
ANOTHER RANDOM UPDATE METHOD: USING BRUTE FORCE
WITHOUT APPLYING PROPOSED METHOD
Congestion occurs when the transient maximum load exceeds that of the initial state.
0.88
0.86
Overflow
Threshold
MAXIMUM LOAD
0.84
0.82
0.8
0.78
0.76
0.74
CONGESTION
INITIAL STATE
0.72
0.7
INITIAL
1)
1
2
3
4
5
TARGET
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
Congestion occurs during transient link cost update
2) In this setup, it is not possible to achieve the optimum routing.
3) Therefore the optimum routing performance is 0% (i.e. 0/24).
32
PROPOSED METHOD TO DETERMINE UPDATABLE LINKS FOR
OPTIMUM ROUTING
INPUT
INFORMATION
Topology information＋
Traffic Information
STEP 1
Identify High Load Link +
Construct Forwarding Tree
STEP 2
Identify Leaf Nodes, then select
Link-Leaf Node, which if updated
will optimize maximum load
STEP 3
The Link then qualifies as Updatable Link
OUTPUT INFORMATION
Optimum Routing Updatable Topology
33
PROPOSED UPDATE METHOD: USING BRUTE FORCE TOGETHER
WITH OUR PRIORITY LINK COST UPDATE METHOD
0.88
Overflow Threshold
MAXIMUM LOAD
0.86
0.84
0.82
INITIAL STATE
0.8
0.78
0.76
0.74
0.72
0.7
INITIAL
1
2
3
4
5
TARGET
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
1) No Congestion Occurs
2) All 24 patterns give Optimum Routing.
3) The Optimum Routing Performance is 100% (i.e. 24/24).
34
PERFORMANCE OF RANDOM METHOD Vs. PROPOSED METHOD
METHOD
CONGESTION
RANDOM
CONGESTION OCCURS
PROPOSED (ULLN)
CONGESTION FREE
OPTIMUM ROUTING
100%
100.00%
80.00%
60.00%
40.00%
20.00%
12.50%
0.00%
RANDOM METHOD
ULLN METHOD
OPTIMUM ROUTING PERFORMANCE
35
CONCLUSION – PART 2
 Presented is a Link Cost Update method to avoid
congestion during planned failures.
 Using this method we can achieve 100% optimum
routing
METHOD
CONGESTION
PERFOMANCE
RANDOM
OCCURS
12.5%
PROPOSED (ULLN ) NOTHING
100%
Paper:- S. Tembo, et. al. “A Method to Avoid Link Overflow During Transient Link Cost Update
in IP Networks”. Planning to Submit for publication to the IEICE Transaction on Communications.
36
SUMMARY
37
SUMMARY
38
Acknowledgements
My gratitude goes to the Doctoral Supervisory
Committee members at Akita University (Japan):
1. Prof. Ken-ichi YUKIMATSU,
2. Prof. Hideo TAMAMOTO,
3. Prof. Akihiro YAMAMURA and
4. Prof. Hitoshi OBARA
for their Academic advise and support.
I would like to sincerely thank NTT for funding the
research.
39
THANK
YOU
40